Speed up md-relaxation test

[thempel]

This PR removes the pre-equilibration steps at test time, and reduces local energy minimizer tolerance. Speedup is from 24s to ~4.5s on my sandbox [edit: from > 660s to 2.8s on CI] for test_mdrelax_integration in tests/test_mdrelax.py.

[github-actions]

Summary

Summary

Generated on:	01/20/2026 - 14:52:25
Parser:	Cobertura
Assemblies:	6
Classes:	25
Files:	25
Line coverage:	68.9% (1855 of 2689)
Covered lines:	1855
Uncovered lines:	834
Coverable lines:	2689
Total lines:	8850
Covered branches:	0
Total branches:	0
Method coverage:	Feature is only available for sponsors

Coverage

src.bioemu - 80.3%

Name	Line	Branch
src.bioemu	80.3%	****
init.py	100%
chemgraph.py	100%
convert_chemgraph.py	70%
denoiser.py	97.8%
get_embeds.py	68.6%
md_utils.py	85.8%
model_utils.py	29.2%
models.py	94.1%
run_hpacker.py	0%
sample.py	50%
sde_lib.py	86.6%
seq_io.py	93.9%
shortcuts.py	100%
sidechain_relax.py	77.2%
so3_sde.py	90.3%
structure_module.py	84.3%
utils.py	59.3%

src.bioemu.colabfold_setup -

Name	Line	Branch
src.bioemu.colabfold_setup	****	****
init.py

src.bioemu.hpacker_setup - 58.8%

Name	Line	Branch
src.bioemu.hpacker_setup	58.8%	****
init.py
setup_hpacker.py	58.8%

src.bioemu.openfold.np - 44%

Name	Line	Branch
src.bioemu.openfold.np	44%	****
protein.py	31.2%
residue_constants.py	60.7%

src.bioemu.openfold.utils - 50.1%

Name	Line	Branch
src.bioemu.openfold.utils	50.1%	****
rigid_utils.py	50.1%

src.bioemu.training - 100%

Name	Line	Branch
src.bioemu.training	100%	****
foldedness.py	100%
loss.py	100%

[sarahnlewis]

LGTM and roughly halves the time to run the whole CI pipeline. Thank you!